Give a few string operator overloads a preference boost.
Intent is to approximate: When selecting +/* candidates, treat it ambiguous until finding a user-given arithmetic-vs-string constraint, such as assigning the result to a string. Once a string interpretation is imposed, prefer an alternative that converts to string as soon as possible.
This description is not directly achievable with the CFA type system. The approximation has the known flaw shown in the string-operator test, where a fairly built-up expression that should be ambiguous is actually defaulting to the string version.
This change is the last of the string-overload reorganizations that the string-operator test was originally meant to illustrate. In Mike's opinion, the resulting state is ideal, except for the just-mentioned flaw.
